Ethernet Speed Evolution Overview
From 2.94Mbps in 1973 to 800G/1.6T today, Ethernet speeds have been accelerating faster than ever, driven primarily by AI/ML workloads. Per the Ethernet Alliance 2026 Roadmap, data centers have evolved from 10G servers to 25G/100G/400G/800G interconnects.
Speed Standards at a Glance
25G: SFP28, 1×25G NRZ, server access. 40G: QSFP+, 4×10G, legacy DC. 50G: SFP56, 2×25G, 5G fronthaul. 100G: QSFP28, 4×25G NRZ, mainstream DC. 200G: QSFP56, 4×50G PAM4, AI back-end. 400G: QSFP-DD/OSFP, 8×50G or 4×100G PAM4, hyperscale. 800G: OSFP/QSFP-DD, 8×100G PAM4, AI clusters. 1.6T: OSFP, 8×200G PAM4/Coherent, next-gen AI.
Signaling Evolution
NRZ (25G/lane and below) → PAM4 (50G/lane and above) → Coherent (100G/lane and above for long-haul).
Selection Guide
Enterprise: 25G servers + 100G spine-leaf. Mid-size DC: 100G QSFP28. Hyperscale: 400G QSFP-DD, migrating to 800G. AI clusters: 400G/800G minimum, 1.6T in 2-3 years.
Langzhi Technology offers SFP28, QSFP28, QSFP-DD optical modules from 25G to 400G.
